A lightweight hybrid deep learning system for cardiac valvular disease classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of mortality.
• Phonocardiogram (PCG) signals offer a cost-effective method for heart sound analysis.
• AI and big data are advancing automated diagnosis of cardiac abnormalities.

Purpose of the Study:

• To develop and evaluate a deep learning model for diagnosing five heart valve conditions using PCG signals.
• To assess the performance of the model with both augmented and non-augmented datasets.
• To compare the model's performance against existing methods on benchmark datasets.

Main Methods:

• A hybrid deep learning architecture combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) was developed.
• The model was trained and validated using the open heart sound dataset and the PhysioNet/Computing in Cardiology 2016 challenge dataset.
• Performance was evaluated using accuracy, F1-score, and Area Under the Curve (AUC).

Main Results:

• For the five-class problem (normal, AS, MR, MS, MVP) on the open heart sound dataset, the augmented model achieved 99.87% accuracy, 99.87% F1-score, and 0.9985 AUC.
• The non-augmented model achieved 98.5% accuracy, 98.501% F1-score, and 0.9978 AUC.
• On the PhysioNet 2016 dataset (two-class problem), the model achieved 93.76% accuracy, 85.59% F1-score, and 0.9505 AUC, outperforming previous works.

Conclusions:

• The proposed CNN-LSTM model demonstrates superior performance in diagnosing heart valve conditions from PCG signals.
• Data augmentation significantly enhances the model's diagnostic accuracy and robustness.
• The system shows potential for early and automated detection of cardiovascular diseases, with future work exploring multimodal integration (PCG and ECG).